- class momaapi.data.ann.AAct(info, entities, ias, tas, atts, rels)#
Class for an atomic action annotation. Atomic actions are unary predicates that actors perform.
id_entity – Entity ID
kind_entity – type of the entity
cname_entity – Entity class name
cid_entity – Entity class ID
start – start time of the atomic action in seconds, relative to the start of the activity video
end – end time of the atomic action in seconds, relative to the start of the activity video
- class momaapi.data.ann.Act(ann, taxonomy)#
Class for an activity annotation. An activity is the coarsest level of annotation, consisting of a series of subactivities that are decomposed into smaller subactivities.
cname – Activity class name
cid – Activity class ID
start – Start time of the activity in seconds
end – End time of the activity in seconds
ids_sact – List of sub-activity IDs
- class momaapi.data.ann.BBox(ann)#
Bounding box in the form of [x, y, w, h]. These are utilized to localize entities.
x – x-coordinate of the top-left corner of the bounding box
y – y-coordinate of the top-left corner of the bounding box
w – width of the bounding box
h – height of the bounding box
- class momaapi.data.ann.Clip(ann, neighbors)#
A clip corresponds to a 1 second/5 frames video clip centered at the higher-order interaction - <1 second/5 frames if exceeds the raw video boundary - Currently, only clips from the test set have been generated
- class momaapi.data.ann.Entity(ann, kind, taxonomy)#
Class of an annotation of an entity. Entities are the building blocks of interactions. They are either human actors or inhuman objects.
id – entity ID
kind – kind of the entity, either “actor” or “object”
cname – class name of the entity
cid – class ID of the entity
bbox – bounding box of the entity
- class momaapi.data.ann.HOI(ann, taxonomy_actor, taxonomy_object, taxonomy_ia, taxonomy_ta, taxonomy_att, taxonomy_rel)#
Class for a higher order interaction. A higher-order interaction, abbreviated as HOI, is a predicate involving two or more entities.
id – HOI annotation ID
time – time of the HOI annotation in seconds, relative to the start of the activity video
actors – list of actor entities involved in the interaction
ias – list of intransitive actions occuring between actors
tas – list of transitive actions occuring between actors
atts – list of attributes that the actor has
rels – list of relationships between entities in the interaction
- class momaapi.data.ann.Metadatum(ann)#
Metadata class for a video. The metadata contains information for videos in the MOMA-LRG dataset, the properties of which are detailed below.
id – Activity ID
fname – File name of the video
num_frames – Number of frames in the video
width – Width of the video resolution
height – Height of the video resolution
duration – Duration of the video in seconds
Get the frame ID given a timestamp in seconds :param time: Timestamp in seconds :type time: float
- class momaapi.data.ann.Predicate(ann, kind, taxonomy)#
Predicate class, representing unary and binary predicates. Predicates are of the form
[src] (cid) [trg], where
srcrefers to the “source entity” performing the action and
trgto the “target entity” who is affected by the source entity.
kind – kind of the predicate
cname – class name of the predicate
id_src – ID of the source entity
id_trg – ID of the target entity
- class momaapi.data.ann.SAct(ann, scale_factor, taxonomy_sact, taxonomy_actor, taxonomy_object, taxonomy_ia, taxonomy_ta, taxonomy_att, taxonomy_rel)#
Class for a sub-activity class annotation. A subactivity is a finer grained level of annotation which refers to a step taken as part of an activity. It is temporallly localized within the activity (that is, it has a start and end time in seconds that are relative to the start of the activity).
cname – Sub-activity class name
cid – Sub-activity class ID
start – Start time of the sub-activity in seconds, relative to the start of the activity video
end – End time of the sub-activity in seconds, relative to the start of the activity video
ids_hoi – List of higher-order interactions
times – Times of higher order interactions inside the video